2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
25 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_quota.h"
37 #include "xfs_utils.h"
38 #include "xfs_trans_priv.h"
39 #include "xfs_inode_item.h"
41 #include "xfs_trace.h"
45 * Define xfs inode iolock lockdep classes. We need to ensure that all active
46 * inodes are considered the same for lockdep purposes, including inodes that
47 * are recycled through the XFS_IRECLAIMABLE state. This is the the only way to
48 * guarantee the locks are considered the same when there are multiple lock
49 * initialisation siteѕ. Also, define a reclaimable inode class so it is
50 * obvious in lockdep reports which class the report is against.
52 static struct lock_class_key xfs_iolock_active
;
53 struct lock_class_key xfs_iolock_reclaimable
;
56 * Allocate and initialise an xfs_inode.
58 STATIC
struct xfs_inode
*
66 * if this didn't occur in transactions, we could use
67 * KM_MAYFAIL and return NULL here on ENOMEM. Set the
68 * code up to do this anyway.
70 ip
= kmem_zone_alloc(xfs_inode_zone
, KM_SLEEP
);
73 if (inode_init_always(mp
->m_super
, VFS_I(ip
))) {
74 kmem_zone_free(xfs_inode_zone
, ip
);
78 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
79 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
80 ASSERT(!xfs_isiflocked(ip
));
81 ASSERT(ip
->i_ino
== 0);
83 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
84 lockdep_set_class_and_name(&ip
->i_iolock
.mr_lock
,
85 &xfs_iolock_active
, "xfs_iolock_active");
87 /* initialise the xfs inode */
90 memset(&ip
->i_imap
, 0, sizeof(struct xfs_imap
));
92 memset(&ip
->i_df
, 0, sizeof(xfs_ifork_t
));
94 ip
->i_update_core
= 0;
95 ip
->i_delayed_blks
= 0;
96 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
104 xfs_inode_free_callback(
105 struct rcu_head
*head
)
107 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
108 struct xfs_inode
*ip
= XFS_I(inode
);
110 kmem_zone_free(xfs_inode_zone
, ip
);
115 struct xfs_inode
*ip
)
117 switch (ip
->i_d
.di_mode
& S_IFMT
) {
121 xfs_idestroy_fork(ip
, XFS_DATA_FORK
);
126 xfs_idestroy_fork(ip
, XFS_ATTR_FORK
);
130 * Only if we are shutting down the fs will we see an
131 * inode still in the AIL. If it is there, we should remove
132 * it to prevent a use-after-free from occurring.
134 xfs_log_item_t
*lip
= &ip
->i_itemp
->ili_item
;
135 struct xfs_ail
*ailp
= lip
->li_ailp
;
137 ASSERT(((lip
->li_flags
& XFS_LI_IN_AIL
) == 0) ||
138 XFS_FORCED_SHUTDOWN(ip
->i_mount
));
139 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
140 spin_lock(&ailp
->xa_lock
);
141 if (lip
->li_flags
& XFS_LI_IN_AIL
)
142 xfs_trans_ail_delete(ailp
, lip
);
144 spin_unlock(&ailp
->xa_lock
);
146 xfs_inode_item_destroy(ip
);
150 /* asserts to verify all state is correct here */
151 ASSERT(atomic_read(&ip
->i_pincount
) == 0);
152 ASSERT(!spin_is_locked(&ip
->i_flags_lock
));
153 ASSERT(!xfs_isiflocked(ip
));
156 * Because we use RCU freeing we need to ensure the inode always
157 * appears to be reclaimed with an invalid inode number when in the
158 * free state. The ip->i_flags_lock provides the barrier against lookup
161 spin_lock(&ip
->i_flags_lock
);
162 ip
->i_flags
= XFS_IRECLAIM
;
164 spin_unlock(&ip
->i_flags_lock
);
166 call_rcu(&VFS_I(ip
)->i_rcu
, xfs_inode_free_callback
);
170 * Check the validity of the inode we just found it the cache
174 struct xfs_perag
*pag
,
175 struct xfs_inode
*ip
,
178 int lock_flags
) __releases(RCU
)
180 struct inode
*inode
= VFS_I(ip
);
181 struct xfs_mount
*mp
= ip
->i_mount
;
185 * check for re-use of an inode within an RCU grace period due to the
186 * radix tree nodes not being updated yet. We monitor for this by
187 * setting the inode number to zero before freeing the inode structure.
188 * If the inode has been reallocated and set up, then the inode number
189 * will not match, so check for that, too.
191 spin_lock(&ip
->i_flags_lock
);
192 if (ip
->i_ino
!= ino
) {
193 trace_xfs_iget_skip(ip
);
194 XFS_STATS_INC(xs_ig_frecycle
);
201 * If we are racing with another cache hit that is currently
202 * instantiating this inode or currently recycling it out of
203 * reclaimabe state, wait for the initialisation to complete
206 * XXX(hch): eventually we should do something equivalent to
207 * wait_on_inode to wait for these flags to be cleared
208 * instead of polling for it.
210 if (ip
->i_flags
& (XFS_INEW
|XFS_IRECLAIM
)) {
211 trace_xfs_iget_skip(ip
);
212 XFS_STATS_INC(xs_ig_frecycle
);
218 * If lookup is racing with unlink return an error immediately.
220 if (ip
->i_d
.di_mode
== 0 && !(flags
& XFS_IGET_CREATE
)) {
226 * If IRECLAIMABLE is set, we've torn down the VFS inode already.
227 * Need to carefully get it back into useable state.
229 if (ip
->i_flags
& XFS_IRECLAIMABLE
) {
230 trace_xfs_iget_reclaim(ip
);
233 * We need to set XFS_IRECLAIM to prevent xfs_reclaim_inode
234 * from stomping over us while we recycle the inode. We can't
235 * clear the radix tree reclaimable tag yet as it requires
236 * pag_ici_lock to be held exclusive.
238 ip
->i_flags
|= XFS_IRECLAIM
;
240 spin_unlock(&ip
->i_flags_lock
);
243 error
= -inode_init_always(mp
->m_super
, inode
);
246 * Re-initializing the inode failed, and we are in deep
247 * trouble. Try to re-add it to the reclaim list.
250 spin_lock(&ip
->i_flags_lock
);
252 ip
->i_flags
&= ~(XFS_INEW
| XFS_IRECLAIM
);
253 ASSERT(ip
->i_flags
& XFS_IRECLAIMABLE
);
254 trace_xfs_iget_reclaim_fail(ip
);
258 spin_lock(&pag
->pag_ici_lock
);
259 spin_lock(&ip
->i_flags_lock
);
262 * Clear the per-lifetime state in the inode as we are now
263 * effectively a new inode and need to return to the initial
264 * state before reuse occurs.
266 ip
->i_flags
&= ~XFS_IRECLAIM_RESET_FLAGS
;
267 ip
->i_flags
|= XFS_INEW
;
268 __xfs_inode_clear_reclaim_tag(mp
, pag
, ip
);
269 inode
->i_state
= I_NEW
;
271 ASSERT(!rwsem_is_locked(&ip
->i_iolock
.mr_lock
));
272 mrlock_init(&ip
->i_iolock
, MRLOCK_BARRIER
, "xfsio", ip
->i_ino
);
273 lockdep_set_class_and_name(&ip
->i_iolock
.mr_lock
,
274 &xfs_iolock_active
, "xfs_iolock_active");
276 spin_unlock(&ip
->i_flags_lock
);
277 spin_unlock(&pag
->pag_ici_lock
);
279 /* If the VFS inode is being torn down, pause and try again. */
281 trace_xfs_iget_skip(ip
);
286 /* We've got a live one. */
287 spin_unlock(&ip
->i_flags_lock
);
289 trace_xfs_iget_hit(ip
);
293 xfs_ilock(ip
, lock_flags
);
295 xfs_iflags_clear(ip
, XFS_ISTALE
);
296 XFS_STATS_INC(xs_ig_found
);
301 spin_unlock(&ip
->i_flags_lock
);
309 struct xfs_mount
*mp
,
310 struct xfs_perag
*pag
,
313 struct xfs_inode
**ipp
,
317 struct xfs_inode
*ip
;
319 xfs_agino_t agino
= XFS_INO_TO_AGINO(mp
, ino
);
321 ip
= xfs_inode_alloc(mp
, ino
);
325 error
= xfs_iread(mp
, tp
, ip
, flags
);
329 trace_xfs_iget_miss(ip
);
331 if ((ip
->i_d
.di_mode
== 0) && !(flags
& XFS_IGET_CREATE
)) {
337 * Preload the radix tree so we can insert safely under the
338 * write spinlock. Note that we cannot sleep inside the preload
341 if (radix_tree_preload(GFP_KERNEL
)) {
347 * Because the inode hasn't been added to the radix-tree yet it can't
348 * be found by another thread, so we can do the non-sleeping lock here.
351 if (!xfs_ilock_nowait(ip
, lock_flags
))
355 spin_lock(&pag
->pag_ici_lock
);
357 /* insert the new inode */
358 error
= radix_tree_insert(&pag
->pag_ici_root
, agino
, ip
);
359 if (unlikely(error
)) {
360 WARN_ON(error
!= -EEXIST
);
361 XFS_STATS_INC(xs_ig_dup
);
363 goto out_preload_end
;
366 /* These values _must_ be set before releasing the radix tree lock! */
367 ip
->i_udquot
= ip
->i_gdquot
= NULL
;
368 xfs_iflags_set(ip
, XFS_INEW
);
370 spin_unlock(&pag
->pag_ici_lock
);
371 radix_tree_preload_end();
377 spin_unlock(&pag
->pag_ici_lock
);
378 radix_tree_preload_end();
380 xfs_iunlock(ip
, lock_flags
);
382 __destroy_inode(VFS_I(ip
));
388 * Look up an inode by number in the given file system.
389 * The inode is looked up in the cache held in each AG.
390 * If the inode is found in the cache, initialise the vfs inode
393 * If it is not in core, read it in from the file system's device,
394 * add it to the cache and initialise the vfs inode.
396 * The inode is locked according to the value of the lock_flags parameter.
397 * This flag parameter indicates how and if the inode's IO lock and inode lock
400 * mp -- the mount point structure for the current file system. It points
401 * to the inode hash table.
402 * tp -- a pointer to the current transaction if there is one. This is
403 * simply passed through to the xfs_iread() call.
404 * ino -- the number of the inode desired. This is the unique identifier
405 * within the file system for the inode being requested.
406 * lock_flags -- flags indicating how to lock the inode. See the comment
407 * for xfs_ilock() for a list of valid values.
423 /* reject inode numbers outside existing AGs */
424 if (!ino
|| XFS_INO_TO_AGNO(mp
, ino
) >= mp
->m_sb
.sb_agcount
)
427 /* get the perag structure and ensure that it's inode capable */
428 pag
= xfs_perag_get(mp
, XFS_INO_TO_AGNO(mp
, ino
));
429 agino
= XFS_INO_TO_AGINO(mp
, ino
);
434 ip
= radix_tree_lookup(&pag
->pag_ici_root
, agino
);
437 error
= xfs_iget_cache_hit(pag
, ip
, ino
, flags
, lock_flags
);
439 goto out_error_or_again
;
442 XFS_STATS_INC(xs_ig_missed
);
444 error
= xfs_iget_cache_miss(mp
, pag
, tp
, ino
, &ip
,
447 goto out_error_or_again
;
454 * If we have a real type for an on-disk inode, we can set ops(&unlock)
455 * now. If it's a new inode being created, xfs_ialloc will handle it.
457 if (xfs_iflags_test(ip
, XFS_INEW
) && ip
->i_d
.di_mode
!= 0)
462 if (error
== EAGAIN
) {
471 * This is a wrapper routine around the xfs_ilock() routine
472 * used to centralize some grungy code. It is used in places
473 * that wish to lock the inode solely for reading the extents.
474 * The reason these places can't just call xfs_ilock(SHARED)
475 * is that the inode lock also guards to bringing in of the
476 * extents from disk for a file in b-tree format. If the inode
477 * is in b-tree format, then we need to lock the inode exclusively
478 * until the extents are read in. Locking it exclusively all
479 * the time would limit our parallelism unnecessarily, though.
480 * What we do instead is check to see if the extents have been
481 * read in yet, and only lock the inode exclusively if they
484 * The function returns a value which should be given to the
485 * corresponding xfs_iunlock_map_shared(). This value is
486 * the mode in which the lock was actually taken.
489 xfs_ilock_map_shared(
494 if ((ip
->i_d
.di_format
== XFS_DINODE_FMT_BTREE
) &&
495 ((ip
->i_df
.if_flags
& XFS_IFEXTENTS
) == 0)) {
496 lock_mode
= XFS_ILOCK_EXCL
;
498 lock_mode
= XFS_ILOCK_SHARED
;
501 xfs_ilock(ip
, lock_mode
);
507 * This is simply the unlock routine to go with xfs_ilock_map_shared().
508 * All it does is call xfs_iunlock() with the given lock_mode.
511 xfs_iunlock_map_shared(
513 unsigned int lock_mode
)
515 xfs_iunlock(ip
, lock_mode
);
519 * The xfs inode contains 2 locks: a multi-reader lock called the
520 * i_iolock and a multi-reader lock called the i_lock. This routine
521 * allows either or both of the locks to be obtained.
523 * The 2 locks should always be ordered so that the IO lock is
524 * obtained first in order to prevent deadlock.
526 * ip -- the inode being locked
527 * lock_flags -- this parameter indicates the inode's locks
528 * to be locked. It can be:
533 * XFS_IOLOCK_SHARED | XFS_ILOCK_SHARED,
534 * XFS_IOLOCK_SHARED | XFS_ILOCK_EXCL,
535 * XFS_IOLOCK_EXCL | XFS_ILOCK_SHARED,
536 * XFS_IOLOCK_EXCL | XFS_ILOCK_EXCL
544 * You can't set both SHARED and EXCL for the same lock,
545 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
546 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
548 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
549 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
550 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
551 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
552 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
554 if (lock_flags
& XFS_IOLOCK_EXCL
)
555 mrupdate_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
556 else if (lock_flags
& XFS_IOLOCK_SHARED
)
557 mraccess_nested(&ip
->i_iolock
, XFS_IOLOCK_DEP(lock_flags
));
559 if (lock_flags
& XFS_ILOCK_EXCL
)
560 mrupdate_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
561 else if (lock_flags
& XFS_ILOCK_SHARED
)
562 mraccess_nested(&ip
->i_lock
, XFS_ILOCK_DEP(lock_flags
));
564 trace_xfs_ilock(ip
, lock_flags
, _RET_IP_
);
568 * This is just like xfs_ilock(), except that the caller
569 * is guaranteed not to sleep. It returns 1 if it gets
570 * the requested locks and 0 otherwise. If the IO lock is
571 * obtained but the inode lock cannot be, then the IO lock
572 * is dropped before returning.
574 * ip -- the inode being locked
575 * lock_flags -- this parameter indicates the inode's locks to be
576 * to be locked. See the comment for xfs_ilock() for a list
585 * You can't set both SHARED and EXCL for the same lock,
586 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
587 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
589 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
590 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
591 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
592 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
593 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_LOCK_DEP_MASK
)) == 0);
595 if (lock_flags
& XFS_IOLOCK_EXCL
) {
596 if (!mrtryupdate(&ip
->i_iolock
))
598 } else if (lock_flags
& XFS_IOLOCK_SHARED
) {
599 if (!mrtryaccess(&ip
->i_iolock
))
602 if (lock_flags
& XFS_ILOCK_EXCL
) {
603 if (!mrtryupdate(&ip
->i_lock
))
604 goto out_undo_iolock
;
605 } else if (lock_flags
& XFS_ILOCK_SHARED
) {
606 if (!mrtryaccess(&ip
->i_lock
))
607 goto out_undo_iolock
;
609 trace_xfs_ilock_nowait(ip
, lock_flags
, _RET_IP_
);
613 if (lock_flags
& XFS_IOLOCK_EXCL
)
614 mrunlock_excl(&ip
->i_iolock
);
615 else if (lock_flags
& XFS_IOLOCK_SHARED
)
616 mrunlock_shared(&ip
->i_iolock
);
622 * xfs_iunlock() is used to drop the inode locks acquired with
623 * xfs_ilock() and xfs_ilock_nowait(). The caller must pass
624 * in the flags given to xfs_ilock() or xfs_ilock_nowait() so
625 * that we know which locks to drop.
627 * ip -- the inode being unlocked
628 * lock_flags -- this parameter indicates the inode's locks to be
629 * to be unlocked. See the comment for xfs_ilock() for a list
630 * of valid values for this parameter.
639 * You can't set both SHARED and EXCL for the same lock,
640 * and only XFS_IOLOCK_SHARED, XFS_IOLOCK_EXCL, XFS_ILOCK_SHARED,
641 * and XFS_ILOCK_EXCL are valid values to set in lock_flags.
643 ASSERT((lock_flags
& (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
)) !=
644 (XFS_IOLOCK_SHARED
| XFS_IOLOCK_EXCL
));
645 ASSERT((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) !=
646 (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
));
647 ASSERT((lock_flags
& ~(XFS_LOCK_MASK
| XFS_IUNLOCK_NONOTIFY
|
648 XFS_LOCK_DEP_MASK
)) == 0);
649 ASSERT(lock_flags
!= 0);
651 if (lock_flags
& XFS_IOLOCK_EXCL
)
652 mrunlock_excl(&ip
->i_iolock
);
653 else if (lock_flags
& XFS_IOLOCK_SHARED
)
654 mrunlock_shared(&ip
->i_iolock
);
656 if (lock_flags
& XFS_ILOCK_EXCL
)
657 mrunlock_excl(&ip
->i_lock
);
658 else if (lock_flags
& XFS_ILOCK_SHARED
)
659 mrunlock_shared(&ip
->i_lock
);
661 if ((lock_flags
& (XFS_ILOCK_SHARED
| XFS_ILOCK_EXCL
)) &&
662 !(lock_flags
& XFS_IUNLOCK_NONOTIFY
) && ip
->i_itemp
) {
664 * Let the AIL know that this item has been unlocked in case
665 * it is in the AIL and anyone is waiting on it. Don't do
666 * this if the caller has asked us not to.
668 xfs_trans_unlocked_item(ip
->i_itemp
->ili_item
.li_ailp
,
669 (xfs_log_item_t
*)(ip
->i_itemp
));
671 trace_xfs_iunlock(ip
, lock_flags
, _RET_IP_
);
675 * give up write locks. the i/o lock cannot be held nested
676 * if it is being demoted.
683 ASSERT(lock_flags
& (XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
));
684 ASSERT((lock_flags
& ~(XFS_IOLOCK_EXCL
|XFS_ILOCK_EXCL
)) == 0);
686 if (lock_flags
& XFS_ILOCK_EXCL
)
687 mrdemote(&ip
->i_lock
);
688 if (lock_flags
& XFS_IOLOCK_EXCL
)
689 mrdemote(&ip
->i_iolock
);
691 trace_xfs_ilock_demote(ip
, lock_flags
, _RET_IP_
);
700 if (lock_flags
& (XFS_ILOCK_EXCL
|XFS_ILOCK_SHARED
)) {
701 if (!(lock_flags
& XFS_ILOCK_SHARED
))
702 return !!ip
->i_lock
.mr_writer
;
703 return rwsem_is_locked(&ip
->i_lock
.mr_lock
);
706 if (lock_flags
& (XFS_IOLOCK_EXCL
|XFS_IOLOCK_SHARED
)) {
707 if (!(lock_flags
& XFS_IOLOCK_SHARED
))
708 return !!ip
->i_iolock
.mr_writer
;
709 return rwsem_is_locked(&ip
->i_iolock
.mr_lock
);
719 struct xfs_inode
*ip
)
721 wait_queue_head_t
*wq
= bit_waitqueue(&ip
->i_flags
, __XFS_IFLOCK_BIT
);
722 DEFINE_WAIT_BIT(wait
, &ip
->i_flags
, __XFS_IFLOCK_BIT
);
725 prepare_to_wait_exclusive(wq
, &wait
.wait
, TASK_UNINTERRUPTIBLE
);
726 if (xfs_isiflocked(ip
))
728 } while (!xfs_iflock_nowait(ip
));
730 finish_wait(wq
, &wait
.wait
);